Date: Nov 28, 2012 6:26 PM
Author: Halitsky
Subject: Thanks for your review of Tables I/II from previous analysis

Thanks for your review of Tables I/II from previous analysis

Thanks for taking the time to do that review.

You wrote:

?re Table I: "% Chg" (percent change?) is not a good measure here.
The analysis itself looks at the simple differences. Percent change,
(or its alternate form, relative difference) is rarely appropriate
when the two values involved can have different signs.?

OK - I see. In any event, I?m hoping that the nature of the results
from the 2nd round of analysis will such as to pass the ?IOTT? in a
very obvious way (unless you?re analysis of the means and average
slopes in my most recent two posts tell you that we can?t use your
first cut on a definition of average slope.

You wrote:

?re Table II: The trick is to compare p(j)*(14-j) to alpha, instead
of comparing p(j) to alpha/(14-j). That way you can simply scan down
and see what you would have to change your chosen alpha to in order
to declare the j'th test "significant". Thus, .005 * (14-3) = .055,
which is twice your chosen alpha (.025) and so should probably not be
called significant.?

Yes ? I?ve recomputed and see exactly what you mean. Don?t know how
rules of construction of Bonferonni tables got garbled on my end since
we used them for the original cust-het-t-tests. Thanks for
straightening me out.

You wrote:

"Also: why .025 instead of .05? Are the p's one-tailed when you
want them to be two-tailed? Such things should be part of the report."

No - they're 2-tailed. I picked .025 just to have something that I
thought would be less arguable than .05.